AI search optimisation checklist: 24 checks for a UK business website

AI-assisted search often combines information from multiple sources. That makes consistent entity representation valuable. The organisation name should be stable across the website and credible external profiles. Service terminology should not change randomly between pages. People who author expert material should have real profiles rather than invented biographies. If the business serves the whole UK remotely, the site should say that instead of manufacturing local offices.

Organisation structured data should describe facts that are actually visible or verifiable: name, canonical website, real contact details when available, real social profiles and areas served. Service schema should connect back to the same organisation entity. Article schema should identify a genuine author when the author is public. Structured data is a classification layer; it is not evidence on its own.

Internal linking contributes to entity clarity as well. A guide about AI search should link to AEO, GEO, technical SEO, digital PR and relevant tools where those relationships genuinely help the reader. Repeatedly inserting the same exact-match anchor everywhere is less useful than a deliberate knowledge graph that follows the subject.

Authority work for AI search should look like good authority work for humans. Publish resources that relevant people would genuinely reference. Use transparent sources. Earn coverage because an asset, opinion, dataset or tool is useful—not because a placement package promises a fixed number of links.

Digital PR and link building should prioritise topical relevance, editorial judgement and the quality of the referring source. Google’s spam policies apply to generative search experiences as well as conventional results, so scaled link schemes, doorway content and manipulative site-reputation tactics create risk rather than durable visibility.

For claims that rely on external facts, cite the primary source where possible. For claims about your own performance, use internal evidence and do not publish results until they have been verified. For expert guidance, show who wrote it and why their experience is relevant. These practices make the site more accountable to a human reader and easier for a machine to corroborate.

There is no universal, stable “position 3 in ChatGPT” metric equivalent to a traditional search ranking. AI answers can vary with wording, context, location, freshness and model behaviour. Measurement therefore needs a panel rather than a single score.

Start with conventional search data: index coverage, impressions, clicks, landing-page performance and qualified conversions. Add a repeatable AI visibility panel using a fixed set of commercially relevant research questions. Record whether the business is named, whether the domain is cited, which competitors are repeatedly surfaced and what sources appear to influence the answer. Use the panel directionally over time rather than presenting it as deterministic market share.

Then connect visibility to the commercial outcome. A citation that never sends qualified demand may still contribute to awareness, but it should not be reported as revenue. A page that gains search visibility but produces low-quality enquiries may need conversion or intent work rather than more traffic. Measurement should decide the next action, not decorate the monthly report.